Usage

Starting a session

Navigate to your project directory and launch the interactive TUI:

cd my-project
interpreter

Or use the short alias:

i

This opens a ChatGPT-like chat interface in your terminal. Type a task in plain English and the agent reads your project, proposes changes, edits files, and runs commands.

Example: adding an endpoint

> add a /health endpoint that returns the build sha

Open Interpreter reads the repository, proposes the file changes, and executes them through the active sandbox after your approval.

Running a one-shot command (non-interactive)

Use exec to run a single task without entering the interactive session:

interpreter exec "summarize the current diff"

Switching models

Inside a session, use the /model command to switch providers or models:

> /model

A picker appears listing configured providers. Select and confirm.

Switching harnesses

Use /harness to change the active agent harness:

> /harness

native
claude-code
claude-code-bare
kimi-cli
qwen-code
deepseek-tui
swe-agent
minimal

Different harnesses emulate the prompt engineering strategies of different coding agents. Pick the one that matches your model for best results.

ACP agent mode (editor integration)

Run Open Interpreter as an Agent Client Protocol agent so editors like Cursor or VS Code can connect to it:

interpreter acp

Sandbox and permissions

Sandbox mode and approval policy are set via flags or the /permissions TUI command. See the sandbox docs for details.